Bandwagon, Underdog, or Projection? Opinion Polls and Electoral Choice in m$ Britain, 1979-1987 B Ian McAllister; Donley T. Studlar The Journal of Politics, Vol. 53, No. 3. (Aug., 1991), pp. 720-741. Stable URL: http://linksjstor.or^^ The Journal of Politics is currently published by Southern Political Science Association. Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http ://www.j stor.org/about/terms .html . JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/spsa.html . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact email@example.com. http : //www . j s tor . org Wed Feb 21 20:35:08 2007 Bandwagon, Underdog, or Projection? Opinion Polls and Electoral Choice in Since the widespread use of opinion polls in British general elections began in the 1950s, there has been continuing controversy concerning their impact on the vote. The bandwagon effect sees voters favoring a party that is doing well in the polls, while the underdog effect predicts that support will go to a party trailing in the polls. There is also the possibility of a projection effect, with voters' expectations conforming to their partisanship. The results pre- sented here, applying logistic regression to "exit poll" survey data collected during the 1979, 1983, and 1987 British general election campaigns, find some evidence of a bandwagon effect in all three elections, but no evidence of an underdog effect and only minor evidence of a projection effect. However, there is a consistent interaction between poll influence and those who decided which way to vote during the election campaign, suggesting that opinion polls can facilitate tactical voting, especially in three-party competition. /Vs campaign polls in Britain have proliferated, there has been an ongoing controversy during the last 30 years over their electoral effects. Such has been their perceived influence on election outcomes that calls for legislation restricting the publishing of opinion polls during a campaign have been fre- quently voiced, a restriction which already exists in countries as diverse as France, South Korea, and Brazil (Worcester 1980; Kavanagh 1981; Whiteley 1986; Norris 1987b). But do campaign polls really affect voting choice and the election out- come? The British evidence, like that from the United States, is conflicting. The main debate concerns whether polls induce a bandwagon or an under- The 1979, 1983, and 1987 Gallup Exit polls were conducted for the BBC, directed by Ivor Crewe, and made available by the ESRC Data Archive at the University of Essex. Neither the original collectors of the data nor the disseminating agency bear any responsibility for the analy- ses and interpretations presented herein. An earlier version of the paper was presented to the Empirical Social Research Conference, University of Queensland, August 1988. Our thanks to participants at the conference and to Bob Darcy for helpful comments; the usual disclaimer applies. Journal of Politics, Vol. 53, No. 3, August 1991 © 1991 by the University of Texas Press Ian McAllister University of New South Wales Donley T. Studlar Oklahoma State University Bandwagon, Underdog, or Projection? 721 dog effect. Briefly, the bandwagon effect has been the label applied to situa- tions where information about majority opinion, widely dispersed in the community, causes some people to alter their opinion to accord with the majority view (Marsh 1984). This change could be a result of a conscious decision on the part of individuals to conform to what they perceive as the political norm; or it could be a subconscious decision, caused by the prevail- ing attitudes and values of the wider social context. By contrast, the under- dog effect suggests that the same information will cause some people to adopt a minority opinion; this change is more likely to be a conscious one, and to reflect sympathy for the minority view (Gartner 1976; Straffin 1977). Alternatively, there is the possibility of a projection effect, that is, indivi- duals project their intended vote onto their election outcome expectations (Lazarsfeld, Berelson, and Gaudet 1948; Berelson, Lazarsfeld, and McPhee 1954; Mendelsohn and Crespi 1970). This paper clarifies some of the dimensions of this continuing controversy in the context of the British general elections of 1979, 1983, and 1987, using polls conducted on, or just prior to, polling day. Specifically, we ask two questions: (1) how has exposure to opinion polls changed between 1979 and 1987?; and (2) for those who vote, do polls produce bandwagon, underdog, or projection effects? In short, we are asking: under what conditions and on whom do polls have an effect during elections? I. Previous Research Both aggregate and individual-level data have been used in investigations of the impact of published campaign polls. Aggregate research has failed to demonstrate a consistent effect of polls on turnout or party choice. In both Britain and the United States there has been a fall in overall turnout over the last quarter century at the very time when published campaign polls were proliferating, prima facie evidence against published polls positively affecting turnout. There has been a great deal of investigation, using both aggregate and survey data, into the effects of election day predictions of presidential win- ners on the behavior of potential voters in the western United States, where polls are still open as the first exit poll and state results from the eastern portions of the country are being broadcast. The first few instances of this phenomenon generated an almost-unanimous academic consensus that knowing the projected or complete results in the other parts of the country did not effect western levels of turnout, the behavior of undecided voters, or the choices of previously decided voters (Fuchs 1965, 1966; Lang and Lang 1968; Mendelsohn 1966; Mendelsohn and Crespi 1970; Tuchman and Coffin 1971; Wolfinger and Linquiti 1981). More recently, however, this con- clusion has been disputed, in light of the 1980 election when NBC News 722 Ian McAllister and Donley T. Studlar declared Ronald Reagan to be the winner of the presidential race on the basis of exit polls several hours before the voting booths closed in the west, followed by President Jimmy Carters concession of defeat before the polls closed as well. Although the issue remains controversial, most of the studies on the 1980 election find media reports of results and/or Carters concession depressing turnout in the west, to the possible detriment of Democratic candidates lower on the ticket, but not apparently producing a bandwagon for the victorious presidential candidate (Epstein and Strom 1981, 1984; Du- bois 1983; Jackson 1983; Ranney 1983; Delli Carpini 1984; "Early Calls ..." 1985; Sudman 1986). Most of the large-scale research on bandwagon-underdog effects has been based on comparing public opinion poll results during campaigns with ballot box results. If the polls underestimate the winning margin, it is claimed that a bandwagon effect has taken place; if they overpredict the winning margin, then supposedly an underdog effect has occurred. This approach has domi- nated bandwagon-underdog research in British elections, where consistent polling overestimation of the winning margin over five general elections, 1959-October 1974, led to claims of an underdog effect (Economist 1974, 1978a, 1979; Worcester 1980, 1983; Kavanagh 1981; Marsh 1984). Subse- quently, however, the 1979 and 1987 general elections did not show such an underestimation. There have been similarly inconsistent results in aggregate analyses of bandwagon-underdog effects in U.S. elections (Gallup and Rae 1940; Mendelsohn and Crespi 1970; Monroe 1975; Asher 1988). One of the major problems bedevilling research on the influence of public opinion polls has been disentangling their effects from the myriad of other variables possibly influencing the outcomes of elections. Under these cir- cumstances, it is hardly surprising that the most convincing demonstrations of bandwagon-underdog effects in the U.S. have been ones based on theo- retical modeling (Simon 1954; Baumol 1957) or small-scale experimental de- signs (Fleitas 1971; de Bock 1976; Navazio 1977; Ceci and Kahn 1982). The first, however, lacks empirical data, the second, generalizability. Until re- cently, larger-scale surveys have never persuasively demonstrated that per- sistent bandwagon-underdog effects exist (Mendelsohn and Crespi 1970; Roshwalb and Resnicoff 1971) although there is stronger evidence for projec- tion effects (Mendelsohn and Crespi 1970). The long and eventful campaigns in the United States make it difficult to demonstrate convincingly that polls influence electoral behavior. Studies of the impact of bandwagon and momentum effects in presidential nomination campaigns (Beniger 1976; Straffin 1977; Patterson 1980; Aldrich 1980; Ad- ams 1984; Bartels 1988) focus more on how media coverage and primary voting results affect the poll standing, strategies, and delegate-gathering of candidates rather than on the independent effect of polls per se. The most Bandwagon, Underdog, or Projection? 723 relevant of these studies for our purposes, Beniger (1976), argues that, in the long run, candidate standing in countrywide opinion polls is a good guide to eventual nomination success, but that there are no bandwagon effects from poll to poll. Two panel studies of voters in the 1980 presidential campaign indicate the presence of bandwagon effects. Controlling for projection effects, Skalaban (1988) finds that there was a small bandwagon effect for Reagan, especially among voters with the weakest prior opinions. Using Noelle-Neumanns (1984) concept of the spiral of silence, voters move in the direction of per- ceived leaders in races for social acceptance reasons rather than tactical con- siderations. Glynn and McLeod (1984) find that voters in 1980 had both an accurate view of trends in the positions of the candidates and a small ten- dency to shift their votes in the direction of the candidate leading toward the end of the race. A promising approach to investigating bandwagon, projection, and under- dog effects is to ask survey respondents if they have heard the results of opinion polls, what the poll result was, and whether their votes have been influenced by this knowledge. One can therefore ascertain whether: (1) re- spondents have a correct view of what opinion polls have been saying (a crucial information question); and (2) to what extent respondents think opin- ion poll results have an impact on voting behavior. In recent years such questions have been posed in "exit" opinion surveys in Britain. The rise of a competitive "third force" in Britain (the Liberals in the 1970s, subsequently the Liberal-Social Democratic Alliance, and now the Liberal Democrats) has fueled interest in the effects of campaign polls. Third-force strategy has been to try to generate a bandwagon effect for themselves through by-election victories, favorable publicity, good poll results, and the encouragement of tactical voting (Rose 1985; Crewe 1986). If it, rather than Labour, would be seen as the principal challenger to the Conservatives, then an electoral breakthrough might occur. Although tactical voting fundamentally depends on the relative standing of the three major parties in particular constituencies, the polls could play a role. Theoretically, opinion poll results might be expected to have more effect in three-party, winner-take-all elections than in two-party competition in winner- take-all elections or in proportional representation systems. Win- ner-take-all elections provide no compensatory seats for losing parties as PR systems do. Two-party competition in a winner-take-all system, however, should encourage potential voters to think their party could win, at least in that particular constituency. On the other hand, three-party competition gives every voter, in effect, two alternative choices to his/her least preferred party. Negative partisanship has been shown to be an increasingly important 724 Ian McAllister and Donley T. Studlar phenomenon in recent British elections (Crewe 1980). If one is anti-Conser- vative, then a strong and persistent Conservative lead in the opinion polls might encourage voting for the party, either Labour or Alliance, best situ- ated to defeat the Conservatives in that particular constituency. On the other hand, being anti-Labour and recognizing a strong Conservative poll lead could generate further support for the Conservatives. In other words, in a three-party situation, one would have the potential for both bandwagon and underdog effects. There is some aggregate evidence that such tactical voting has occurred in Britain (Spafford 1972; Cain 1978). Furthermore, with its recent persistent three-party competition, Britain provides an especially good test for these hypotheses, in contrast to two-party competition in the United States. Interestingly enough, the U.S. election in which there is greatest evidence of polls affecting election results, that of 1980, did have a significant third-party candidacy, that of John Anderson. The most intensively studied campaign polls in Britain, those of the 1983 election, have produced conflicting results. While all of the three published analyses (Rose 1985; Crewe 1986; Whiteley 1986) of the 1983 BBC-Gallup exit poll results agree that the Alliance benefited from a small underdog effect, they agree on little else. Rose and Crewe indicate the findings are tentative; Whiteley argues that there is "compelling evidence" of a signifi- cant opinion poll effect. Furthermore, Whiteley finds that the Conservatives were less harmed by the Alliance bandwagon than was Labour, another view at odds with the other two. Rose and Whiteley find 5% claiming to be influenced by the polls, Crewe only 3%. The number of respondents on which Whiteley bases his strongly-worded conclusions is more than one- third larger (124 as against 90) than Crewe's. Other data (Rose 1985; Worces- ter 1980) indicate that voters see polls mainly as a harmless diversion, a large proportion have difficulty in remembering the specifics of poll results, and few are directly influenced by them. A close comparative analysis of the 1979, 1983, and 1987 British general elections should provide a good test of the bandwagon, underdog, and pro- jection hypotheses. The 1979 campaign opened with a substantial Conser- vative poll lead (approximately 10%) which shrank as the campaign pro- gressed. Opinion polls in the 1983 election campaign consistently showed the Conservatives with an overwhelming (10% to 20%) lead. The 1987 con- test was closer, with the Conservatives maintaining opinion poll leads of generally small dimensions (8% to 11%) and many observers crediting La- bour with "winning" the campaign, if not the election. Skalaban (1988) sug- gests that polls should have more effect on voters when elections are close, but not so close as for the cues from polls to be ambiguous. Similarly, Cain (1978) argues that tactical voting should be most evident when a race is ex- tremely close. By this reasoning, any observable polls effect should be Bandwagon, Underdog, or Projection? 725 greater in 1979 and 1987 than in 1983. But perhaps neither 1979 nor 1987 was close enough. II. Polls and Election Outcomes As opinion polls have become more numerous during British election campaigns, public exposure to them has also increased. During the 1979 general election campaign, 68% of voters reported having heard of the re- sults of an opinion poll. In 1983, this proportion remained stable at 67% but increased to 74% in 1987. Do the majority of voters who have heard an opinion poll result consider that it has influenced their own political behav- ior? The evidence from the three surveys shows that the proportion of res- pondents claiming such an effect has doubled over the course of the three elections. In 1979 and 1983, 4% said that they had been influenced by a poll, rising to 9% in 1987. Table 1 shows that, in terms of the party breakdown, the Liberals (in 1979) and the Alliance (in 1983 and 1987) have had a consis- tent advantage among those who reported being influenced by the polls. Indeed, in 1983, the Alliance won support from more than half of this group. Labour also gained votes from this group in 1979 and 1987, but not in 1983, while Conservative voters were more numerous among those not reporting being influenced by the polls. Among those who report having heard the results of an opinion poll, the majority are able to correctly identify the leading party and to evaluate the size of its lead (table 2). In the 1979 election, just more than half of Conser- vative and Labour voters and just under half of Liberal voters correctly iden- tified the Conservatives as leading the polls, the vast majority of them con- sidering the lead to be "small" rather than "large." However, a significant minority — in the case of Liberal voters, 48% — thought that Labour had a "small" lead. In fact, only one of the 28 nationwide polls conducted during the course of the campaign produced a Labour lead, and among the nine polls conducted in the last week of campaigning, the average Conservative lead was 5 percentage points (Rose 1981). There' was less ambiguity during the 1983 and 1987 election campaigns. In both elections, all but a handful of voters accurately identified a Conser- vative lead, of varying proportions: in 1983, around nine out of 10 voters thought that the Conservatives had a "large" lead. In the more closely fought 1987 election, there are clearly partisan differences, with Conservative vot- ers being more likely to see their party's lead as "large," and Labour voters more likely to see the Conservative lead as "small." These results would tend to give prima facie evidence for the existence of a bandwagon effect: the greater the perceptions that the Conservatives had a large lead, then the greater the likelihood of voting for that party. How- w h-l PQ 00 i— I I 05 W H O > O C0 H-l H-l O Pli g 2 E O o w u £ w o H-l Q W & o s H-l w CD X -0 ^ cS o c c o « -a 3 .S « a 5 O 2g S 8 3 8 8 2 s? 8 8 O a a? 5 I * If a? 2 *( J) H ^ a =3 * ° ^ ^3 ^ W I 2 a 2 > ^ o O £ | cj si « §1 S SP^ ^ ^ hS * G hS 05 "2 ^ « s ^ a> x ii ttf fl CJ fi Q) £ .5 J3 3 m ^ o w s C/3 a lo a o CJ C o CJ c o CJ S3 H LO H (M (M CM rH rH © H s I> © © © CO s s CO O O O CO CO 05 lO t O CM m lo <o 10 o Tt< 8 CO rH 00 © 00 S S § 8 8 00 - 3 2 3 s s 8 g 8 00 CM i— I 05 a .o '3 a O o > $ I 1 | (5 is & § b b rt 3 3 w e s ^ J J3 o g)^ i g g> e 3 £ cr £ 3 9 J C/3 W C/3 J c/3 I o Q 3. 3 J 728 Ian McAllister and Donley T. Studlar ever, it could be argued that a decision to vote for a particular political party also affects how a voter sees their party's position in the polls: having opted for the party, the voter may thereafter wish their choice to be vindicated by the polls and interpret them accordingly, a projection effect (Mendelsohn and Crespi 1970). Thus, the results in table 2 may simply reflect the fact that Conservative voters are projecting their political choice onto the opinion polls. From this perspective, the only data which could address this implied question of causality are panel data, which interview the same respondents at two or more points in time (see Skalaban 1988). However, there are few panel surveys available for British elections, and none which deal with the issue of opinion poll influence. Even if they were available, panel data would present other problems. Repeated interviewing of the same respondents biases results in favor of greater interest in the election (Worcester 1980; Yalch 1976; Kraut and McConahay 1973; Clausen 1968). An alternative so- lution is to estimate the reciprocal paths between the variables of interest, using a simultaneous equation model; using this model, we can measure the reciprocal paths between party and poll lead, and between party and self- reported influence by polls. However, with some exceptions, data on politi- cal behavior are rarely sufficiently robust to permit reliable simultaneous equation estimates (Asher 1983; Berry 1984). 1 An alternative methodology is to overcome at least some of these difficul- ties by utilizing interaction terms in the multivariate equations. While this approach still uses cross-sectional data, it has significant advantages over a noninteraction model. First, if a projection effect were taking place, by which voters projected their partisan preferences onto the size of the opinion poll lead, then an interaction term should measure this effect. Secondly, following Skalaban (1988), we might also expect that those who left their decision on how to vote until late in the campaign would be more influenced by opinion polls — a "campaign polls" effect. Once again, an interaction term should be able to measure this effect if it is present. In the multivariate analyses in tables 3 to 5, dummy variable interaction terms are used to test these hypotheses. III. Testing the Multivariate Model The multivariate models estimated in tables 3 and 4 are designed to mea- sure the bandwagon and underdog hypotheses, and therefore include voters' perceptions of which party they thought was leading in the polls (as shown in tables 1 and 2). Since the distribution of perceived poll leads varied be- tween 1979, when sizeable proportions of the electorate chose one of three Analyses using this technique largely replicated the results shown here. However, because the first-stage equations were not particularly powerful, we have not included these results here. Bandwagon, Underdog, or Projection? 729 options, and 1983 and 1987, when almost all voters thought the Conserva- tives were ahead and differed only about the magnitude of the lead, the models are shown in two separate tables. Both models include variables to measure campaign effects, judged by whether or not voters reported being influenced by the polls in making their voting decision, and whether their decision was reached early or late in the election campaign. Since we would expect that the complex range of factors that affect vote would be closely linked to partisanship, this is also controlled for in all of the multivariate models. 2 Testing for the possibility of a projection effect in the polls presents par- ticular measurement problems. The projection effect hypothesizes that vot- ers will alter their perceptions of the party leading in the polls to conform to their own partisan preferences. As we have already noted, this question can only be measured definitively by panel data. However, one way to partially test the hypothesis is to cross-classify strength of partisanship with the size of the poll lead. If, as the projection hypothesis predicts, voters alter their perception of the poll lead to conform to their partisanship, we might expect that strong partisans would be more likely to see their party leading in the polls, compared to their counterparts with weaker partisan leanings, other things — including the direction of partisanship — being equal. This possibil- ity is tested in table 5 in the 1983 and 1987 elections; unfortunately, a ques- tion measuring strength of partisanship was not included in the 1979 survey and it has therefore been excluded from this analysis. In all of the models, the dependent variable is dichotomous and for that reason logistic regression techniques are used, which show logit coefficients, together with their respective standard errors. 3 To facilitate interpretation of the results, however, we also include the estimated percentage probabilities of the variable in question. These probabilities are analogous to partial OLS coefficients in a dummy variable regression, and they show the percentage probability of voting for a party given the particular attribute in question. They are estimated by evaluating the logistic regression at the grand mean; full details of all the methods and the variables used in the analyses are given in the appendix. The results in table 3 and table 4 show that in each of the three elections, voters who thought that the Conservatives had a large rather than a small 2 In preliminary analyses, several social structural control variables (mainly age, gender, oc- cupation, home ownership, and trade union membership) were included in order to partial out potentially confounding effects. However, since they had little impact on the size or statistical significance of the independent variables of interest in the analyses, they were eventually ex- cluded on the grounds of parsimony. 3 The decision on which category to exclude among the dummy variables was generally based on size, with the category which was least numerous within the population being chosen for exclusion. in I CO w W H O > w S3 H Q Eif U w E> En o Pli W o Pli o o U I I CO CO 8 00 I I S3 G ^ I I c3 S I I I O 00 7 17 o o CO <N I i " co 2 co o O U 8 I I © <N C\j CO <N 00 1 00 I i 8 3 £^ e £ c in ■i -§1 a be B O § 2 c| o * U V) «5 a) is "3 ^ II 33 •i s 33 1° 1 O bC c s H T3 T3 Q Q T3 T3 C 3 3 'Z w 53 53 n C g w "5 J 1 Oh a) 3 2 c 8 U J J bC 5 3 o 2 5 « be a? .. o o a .2 X Bandwagon, Underdog, or Projection? 731 lead (or, in the case of 1979, a Conservative lead as opposed to a Labour lead or a neck-in-neck contest) were significantly more likely to vote for the party and significantly less likely to vote Labour, net of other things. For example, in 1979 the estimated probabilities suggest that voters were 18% more likely to vote for the Conservatives and 14% less likely to vote Labour, net of other things, including the direction of partisanship. This is evidence of a band- wagon effect. In contrast to the evidence for a bandwagon effect between the Conser- vative and Labour parties, there is no clear pattern among Alliance voters in 1983 and 1987 or among Liberal voters in 1979. In none of the three elec- tions are the coefficients statistically significant and the estimated probabili- ties are either zero (as in 1983) or trivial. There is, therefore, little sign of an underdog effect in these three elections. Overall, the evidence points to- ward the existence of a bandwagon effect. There is also substantial evidence in tables 3 and 4 for the existence of a "campaign polls" effect. In the 1983 and 1987 elections, those who made up their minds on how they would vote late in the campaign, and who reported being influenced by the polls, were significantly less likely to vote either Conservative or Labour. Although the proportion of voters in this category is relatively small, the estimated probabilities show large effects, particularly in 1983. For example, in that election, those who were influenced by the polls but decided on their vote before the campaign began were 24% more likely to vote Conservative than those who reported being influenced by the polls but who decided on their vote while the campaign was in full swing. This suggests that they made a rational decision on how to vote, based on the poll results. By contrast, in the 1979 election, when the outcome was more evenly balanced and the Conservatives were in opposition, the pattern is reversed. However, the real beneficiary of the "campaign polls" effect would appear to be the Liberals and their successors, the Alliance. In each election, those who delayed their decision on how to vote until the campaign was in prog- ress were significantly more likely to vote for the Liberals or the Alliance, net of other things. Once we separate these late deciders into those who reported being influenced by the polls and those who did not, the results indicate that the former were significantly more likely to vote Liberal or Alliance. It is, however, important to place this "campaign polls" effect in perspective. First, as we have already mentioned, the proportion influenced by the polls is small; even in 1987, this amounted to less than 10% of the respondents, so that the potential pool of voters open to this type of persua- sion is negligible compared to the electorate as a whole. Second, there is considerable controversy (Crewe 1982, 1986; Rose 1985; Whiteley 1986) over whether the small number of voters citing this are to be considered an accurate figure or merely the tip of a larger iceberg. As Norris (1987a) notes, w PQ 00 c CO 00 W S3 H Q Eif U E3 En o Pli o Pli 8 b ^3 o G5 oq oq I I I S Si I I Srt CO 8 8 8 I I 1 in I S3 CO <N I I I " I CO 05 ?8 ^ O i~H <N -H C\| 8 § oq i—i co io oq I 05 M< § 3 i— J I t4_ c o o S H C 4) 4) V 12 3 Q Si & ^3 T3 T3 CO 00 T3 O O > T3 T3 o ~ a •S ^ S 3 ,2 9^3 u 3 S Si I I 3 J 5 Q e o U ^ 2 ^ 3 § G5 I CO 1— I I I G5 oj <NJ CO C CO 3 1— I I S g3 3 a co £ CD CD O o H3 U CJ « a; , — i H Q C <L> <L> a see CD Q I I I I CO I ft CO <~H I a i K 1 = 31 SIS 3 3 >-" S i-i cs 3 w t >2 g & a i "■8 i-il £8 33 c o O ^3 p s 3 < > 3 ■4-1 V} ^ C ■3 o O ' O 1 CD ^ ^ > O " ^ OO 2 J) H If 11 «3 15 f— i c 1 1 O <L> -^3 c£ ^3 £ W3 * <L> 5 cq oq I I 8 o r co ©q I I W ffl oo i— ( CO oo G5 w H O > w S3 H O S3 8 J c/3 to O o i-H I 3 8 S io ©q " I H < Oh W o Cm to O 8 2 s $ 3 2 00 <~h ©q o oS ^ r oq I Oh O 4) Set «3 .> ?3 •52 -12 Oh *■£ Oh S 1 1 ^ Oh - <D G O J 2u bo £ C Sh HH J_ W J_ N-/ q =3 ^ o * o * Oh 3 2 IJ o o G O P~ 3 8-1 ^3 C o (J I I I I si i \ «0 r 28 CO I I " i 7 (S I io 3 CD f 5 j "3 *Q 0) - j la l * Oh ^ s se - £ - 1 1 .p Pli n pn n ^3 Oh <U 2 u CD z 5 = a a .8 J c o •5 ^3 736 Ian McAllister and Donley T. Studlar much campaign activity, including opinion polls, may be reinforcing voter behavior rather than changing it. To what extent was there a projection effect in any of the three elections, with voters altering their views of the party leading in the polls to match their own partisan preferences and casting their vote accordingly? Using the interaction between strength of partisanship and perceived poll lead sug- gests that there were projection effects which benefited the Conservatives in both elections analyzed in table 5. In 1983, strong partisans who thought there was a large Conservative poll lead were 12% more likely to vote Con- servative, compared to strong partisans who considered that the Conserva- tive poll lead was small. However, in both elections the effects are relatively minor and neither reach statistical significance at the 1% level. 4 While there is evidence of projection effects, their impact on the vote is negligible. IV. Conclusion Over the last decade, there has been much academic commentary on par- tisan dealignment and electoral volatility in Britain (Sarlvik and Crewe 1983; Crewe 1985; Norris 1987b). With the loosening of the bonds between parties and the electorate, more people are potentially available for influence by campaign effects, such as opinion polls, than ever before. Our results indi- cate that part of that potential is being realized. There is a consistent pattern of apparent bandwagon effects for the leading party through voters' percep- tions of the size of the party lead. But the picture is also a complex one and in at least some instances, there are also small projection effects. In other words, it is possible that many voters who displayed a bandwagon effect may well have voted for the party anyway, since they were projecting their partisanship onto the size of the Conservatives' poll lead. 5 However, our results show no evidence for an underdog effect in any of the three elections analysed here. Our results therefore conflict with those of Whiteley (1986) in that we do not find "com- pelling evidence" of a significant underdog effect in 1983. While our results are also at odds with Rose s (1985) and Crewe s (1986) bivariate findings of a small underdog effect in 1983, we agree with their caution in attributing large voter shifts to opinion polls. Our findings, moreover, do not provide strong evidence for the "spiral of silence" interpretation of public opinion and voting, whereby people tend to conform to what they see as dominant political opinion in their reference group (Noelle-Neumann 1977, 1984; Glynn and McLeod 1984), especially if it conforms with a perceived trend (Marsh 1984). Faced with persistent poll 4 In 1983 the f-value was 1.648, in 1987 it was 0.748. 5 For the reasons mentioned in footnote 3, we were not able to integrate the two sets of variables into one explanatory model and test this hypothesis directly. Bandwagon, Underdog, or Projection? 737 results that demonstrate a lead in popularity for one party, some of the elec- torate tends to shift in that direction. But unlike Noelle-Neumann (1984), we did not find this to be a large phenomenon. Perhaps the relative dynam- ics of this depends on reference groups more immediate to the voter (Field 1983; Noelle-Neumann 1977), but in Britain polls themselves are an increas- ingly important reference group in a more fragmented society. Other studies addressing the spiral of silence hypothesis with countrywide opinion poll data (Webb and Wybrow 1986; Darcy and Laver 1990) have also cast doubt on its validity. The effects of polls on the two trailing parties (Labour and Liberals /Alli- ance) provide some interesting contrasts. Alliance supporters are much more attentive to poll results; the problematical nature of Alliance support leads them to search for information on how the parties stand. For example, in 1983, 72% of Alliance voters reported having heard the results of an opinion poll, compared to 57% of Labour voters and 67% of Conservatives. Only in 1983 did poll results appear to generate Alliance support, based on self- reported influence by the polls. Since Alliance (and, in 1979, Liberal voters) were more likely to make up their minds late in the campaign, the opinion polls would appear to interact with the actual timing of the decision. In other words, opinion polls provide an important reference for third-party voters who are defecting from the two major parties. In comparison to Alliance voters, Labour supporters are less attentive to opinion polls, but even they are increasingly using polls as a reference. For members of these two groups who are exposed to polls, the impact on their final voting choice has differed over the course of the three elections. As- sessing polls as showing a large Conservative lead has tended to depress the Labour vote in the 1980s but has had little impact on the Alliance. This suggests that third-party support has been firmer in the elections of the 1980s than previously, as others (Crewe 1987) have shown. On the other hand, Labour support has become less firm in the 1980s; exposure to opinion polls negative for Labour may have contributed to the decline of the Labour vote. We are unable to assess systematically what effects second-place assessments have on relative support for Labour and the Alliance. Lacking this information, as well as data on relating the general "exit" poll findings to particular constituencies, we cannot fully evaluate the tactical voting argument. But we can say that, before the fracture of the Alliance in 1988, Labour and Alliance supporters were converging in their relationship to opinion polls. Election outcomes are decided by many things. Polls are just one, but they are not unimportant. Our analysis shows polls influence voter behavior through a bandwagon effect although in some circumstances there may also be a small projection effect. Although our evidence does not predate 1979, saturation polling does seem to have had significant political effects on the 738 Ian McAllister and Donley T. Studlar public, but not always in predictable ways. In a closer electoral contest than those examined here, we might expect even greater attentiveness to opinion polls and possibly larger effects. But if there is not a significant third-party challenge to Labour and the Conservatives in the next British general elec- tion, there may be less opportunity for tactical voting; polling effects may be reduced, ceteris paribus. None of this analysis should be construed as an argument for restricting published polls during campaigns. As has often been pointed out, such res- trictions would not extend to "private polls" done for the parties (Gallup and Rae 1940; Worcester 1980; Kavanagh 1981). In general, the behavior of pol- iticians may be more influenced by polls than the behavior of the electorate (Mendelsohn and Crespi 1970). For one thing, the very date of the general election is usually chosen by the prime minister on the basis of private and published polls, as well as other factors (Worcester 1983; Rose 1985). What the elitist argument about banning or restricting the publication of opinion polls during campaigns amounts to, then, is distrust of the voters. The voters could reply that public opinion polls are too important to be left to the politicians. Manuscript submitted 25 August 1989 Final manuscript received 17 September 1990 Technical Appendix Data. The data are the 1979, 1983, and 1987 Gallup Exit polls, conducted for the BBC (N = 2,434, 4, 146, and 4,886, respectively). All three were quota surveys representative of all eligible voters aged 18 or older in England, Scotland, and Wales. 6 All the surveys were conducted on either polling day or the day immediately prior to polling day and as such, are the closest polls available to the actual election result. 7 The 1987 survey was less accurate in predicting the result of the election: while the 1979 and 1983 surveys pre- dicted the results for the major parties within 1%, the 1987 survey was in error by 3%. For that reason, the 1987 survey has been additionally weighted to the actual election result.* In the logistic regression equations reported in table 3 to table 5, the analyses are restricted to respondents who 6 The surveys are called "exit polls" because of their proximity to the general election, not because they are conducted outside polling stations. They therefore poll intending non voters as well as intending voters. All responses are based on face-to-face interviews. 7 To exmaine the possible effects of the campaign, the analyses were replicated for those interviewed on polling day (always a Thursday) and those interviewed on the Wednesday prior to polling day. The results showed no substantive difference between the groups, and for that reason the analyses are presented for the sample as a whole. 8 All the surveys are weighted, with the weighted N being adjusted back to the true N to leave significance tests unaffected. Bandwagon, Underdog, or Projection? 739 had seen an opinion poll and reported voting. In the 1983 and 1987 analyses, a small number of respondents who reported anything other than a small or large Conservative lead are excluded, and in table 5 those who did not an- swer the strength of partisanship question are also excluded. Method. The analysis relies on logistic regression techniques since the dependent variables are all dichotomous (Hanushek and Jackson 1977). The independent variables figuring in the logistic regression analyses are all dummy variables, scored zero or one. To facilitate interpretation of the lo- gistic results, estimated probabilities are shown for the independent vari- ables. These are calculated by evaluating the logistic coefficient at the grand mean; this, in effect, makes a linear estimate of the logistic coefficient. As such, these estimated probabilities should be used only as a general guide to interpreting the analyses, not as a definitive result. References Adams, William C. 1984. "Media Coverage of Campaign 84: A Preliminary Report." Public Opinion 7 (February-March): 48-50. Aldrich, John H. 1980. Before the Convention. Chicago: University of Chicago Press. Asher, Herbert B. 1983. Causal Modelling. Beverly Hills, CA: Sage. Asher, Herbert B. 1988. Polling and the Public. Washington: CQ Press. Bartels, Larry M. 1988. Presidential Primaries and the Dynamics of Public Choice. Princeton: Princeton University Press. Baumol, William J. 1957. "Interactions between Successive Polling Results and Voting Inten- tions." Public Opinion Quarterly 21 : 318-23. Beniger, James R. 1976. "Winning the Presidential Nomination: National Polls and State Pri- mary Elections, 1936-1972." Public Opinion Quarterly 40:22-38. Berelson, Bernard R., Paul F Lazarsfeld, and William N. McPhee. 1954. Voting. Chicago: University of Chicago Press. Berry, William D. 1984. Nonrecursive Causal Models. Beverly Hills, CA: Sage. Cain, Bruce. 1978. "Strategic Voting in Britain." American Journal of Political Science 22:639-55. Ceci, Stephen J., and Edward L. Kahn. 1982. "Jumping on the Bandwagon with the Underdog: The Impact of Attitude Polls on Polling Behaviour." Public Opinion Quarterly 46:228-42. Clausen, Aage R. 1968. "Response Validity: Vote Report." Public Opinion Quarterly 32: 588-606. Crewe, Ivor. 1980. "Negative Partisanship: Some Preliminary Ideas Using British Data." Paper presented to European Consortium for Political Research, Florence. Crewe, Ivor. 1982. '"Improving, But Could Do Better:' The Media and the Polls in the 1979 General Election." In Political Communications: The General Election Campaign of 1979, ed. Robert M. Worcester and Martin Harrop. London: Allen and Unwin. Crewe, Ivor. 1985. "Great Britain." In Electoral Change in Western Democracies, ed. Ivor Crewe and David Denver. New York: St. Martins Press. Crewe, Ivor. 1986. "Saturation Polling, the Media and the 1983 Election." In Political Com- munications: The British General Election Campaign of 1983, ed. Ivor Crewe and Martin Harrop. Cambridge: Cambridge University Press. 740 Ian McAllister and Donley T. Studlar Crewe, Ivor. 1987. "What's Left for Labour: An Analysis of Thatchers Victory." Public Opinion 10 (July/ August): 52-56. Darcy, R., and Michael Laver. 1990. "Referendum Dynamics and the Irish Divorce Amend- ment." Public Opinion Quarterly 54: 1-20. de Bock, Harold. 1976. "Influence of In- State Election Poll Reports on Candidate Preference in 1972." Journalism Quarterly 53:457-62. Delli Carpini, Michael. 1984. "Scooping the Voters? The Consequences of the Networks' Early Call of the 1980 Presidential Race." Journal of Politics 46:866-85. Dubois, Philip L. 1983. "Election Night Projections and Voter Turnout in the West: A Note on the Hazards of Aggregate Data Analysis." American Politics Quarterly 11 : 349-64. "Early Calls of Election Results and Exit Polls: Pros, Cons, and Constitutional Considerations." 1985. Public Opinion Quarterly 49: 1-18. Economist. 1974. "Do Polls Affect Voting?" 12 October. Economist 1978. "Do Polls Change Votes?" 10 June. Economist. 1979. "Polls Watcher Wanted." 12 January. Epstein, Laurily K., and Gerald Strom. 1981. "Election Night Projections and West Coast Turnout." American Politics Quarterly 9:479-91. Epstein, Laurily K., and Gerald Strom. 1984. "Survey Research and Election Night Projec- tions." Public Opinion (February/ March): 48-50. Field, Mervin D. 1983. "Political Opinion Polling in the United States of America." In Public Opinion Polling: An International Review, ed. Robert M. Worcester. New York: St. Martin's Press. Fleitas, Daniel W. 1971. "Bandwagon and Underdog Effects in Minimal Information Elec- tions." American Political Science Review 65:434-38. Fuchs, Daniel A. 1965. "Election Day Newscasts and Their Effect on Western Voter Turnout." Journalism Quarterly 42 : 22-28. Fuchs, Daniel A. 1966. "Election-Day Radio-Television and Western Voting." Public Opinion Quarterly 30:226-36. Gallup, George, and Saul Forbes Rae. 1940. "Is There a Bandwagon Vote?" Public Opinion Quarterly 4:244-49. Gartner, M. 1976. "Endogenous Bandwagon and Underdog Effects." Public Choice 25:83-139. Glynn, Carroll J., and Jack M. McLeod. 1984. "Public Opinion du Jour: An Examination of the Spiral of Silence." Public Opinion Quarterly 48:731-40. Hanushek, Eric A., and John E. Jackson. 1977. Statistical Methods for Social Scientists. New York: Academic Press. Jackson, John E. 1983. "Election Night Reporting and Vote Turnout." American Journal of Political Science 27:615-35. Kavanagh, Dennis. 1981. "Public Opinion Polls." In Democracy at the Polk, ed. David Butler, Howard R. Penniman, and Austin Ranney. Washington: American Enterprise Institute. Kraut, R. E., and J. B. McConahay. 1973. "How Being Interviewed Affects Voting: An Experi- ment." Public Opinion Quarterly 37:398-406. Lang, Kurt, and Gladys E. Lang. 1968. Voting and Nonvoting. Waltham, MA: Blaisdell. Lazarsfeld, Paul F., Bernard Berelson, and Hazel Gaudet. 1948. The Peoples Choice. 2d ed. New York: Columbia University Press. Marsh, Catherine. 1984. "Back on the Bandwagon: The Effect of Opinion Polls on Public Opin- ion." British Journal of Political Science 15:51-74. Mendelsohn, Harold. 1966. "Western Voting and Broadcasts of Results on Presidential Election Day." Public Opinion Quarterly 30:212-25. Mendelsohn, Harold, and Irving Crespi. 1970. Polk, Television and the New Politics. Scranton: Chandler. Monroe, Alan D. 1975. Public Opinion in America. New York: Dodd Mead. Bandwagon, Underdog, or Projection? 741 Navazio, Robert. 1977. "An Experimental Approach to Bandwagon Research." Public Opinion Quarterly 41:217-25. Noelle-Neumann, Elizabeth. 1977. "Turbulences in the Climate of Opinion: Methodological Applications of the Spiral of Silence Theory." Public Opinion Quarterly 41 : 143-58. Noelle-Neumann, Elizabeth. 1984. The Spiral of Silence. Chicago: University of Chicago Press. Norris, Pippa. 1987a. "Four Weeks of Sound and Fury: The 1987 British Election Campaign." Parliamentary Affairs 40 : 458-67. Norris, Pippa. 1987b. Volatility in By-elections. Glasgow: University of Strathclyde Studies in Public Policy No. 163. Patterson, Thomas E. 1980. The Mass Media Election. New York: Praeger. Ranney, Austin. 1983. Channels of Power. New York: Basic Books. Roshwalb, Irving, and Leonard Resnicoff. 1971. "The Impact of Endorsements and Published Polls on the 1970 New York Senatorial Election." Public Opinion Quarterly 35:410-14. Rose, Richard. 1981. "Towards Normality: Opinion Polls in the 1979 Election." In Britain at the Polls, 1979, ed. Howard R. Penniman. Washington, DC: American Enterprise Institute. Rose, Richard. 1985. "Opinion Polls as Feedback Mechanisms: From Cavalry Charge to Elec- tronic Warfare." In Britain at the Polls, 1983, ed. Austin Ranney. Durham: Duke University Press. Sarlvik, Bo, and Ivor Crewe. 1983. Decade of Dealignment. New York: Cambridge. Simon, Herbert A. 1954. "Bandwagon and Underdog Effects and the Possibility of Election Predictions." Public Opinion Quarterly 18:245-53. Skalaban, Andrew. 1988. "Do the Polls Influence Elections? Some 1980 Evidence." Political Behavior 10:136-50. SpafFord, Duff. 1972. "Electoral Systems and Voters Behavior." Comparative Politics 5: 129-34. Straffin, Philip D. 1977. "The Bandwagon Curve." American Journal of Political Science 21:695-709. Sudman, Seymour. 1986. "Do Exit Polls Influence Voting Behaviour?" Public Opinion Quar- terly 50:331-39. Tuchman, Sam, and Thomas E. Coffin. 1971. "The Influence of Election Night Television Broadcasts in a Close Election." Public Opinion Quarterly 45:315-26. Webb, Norman, and Robert Wybrow. 1986. "The Spiral of Silence: A British Perspective." In Political Communications: The British General Election Campaign of 1983, ed. Ivor Crewe and Martin Harrop. Cambridge: Cambridge University Press. Whiteley, Paul. 1986. "The Accuracy and Influence of the Polls in the 1983 General Election." In Political Communications: The British General Election Campaign of 1983, ed. Ivor Crewe and Martin Harrop. Cambridge: Cambridge University Press. Wolfinger, Raymond, and Peter Linquiti. 1981. "Tuning In and Turning Out." Public Opinion 4 (February/March): 56-60. Worcester, Robert M. 1980. "Pollsters, the Press, and Political Polling in Britain." Public Opin- ion Quarterly 44 : 548-66. Worcester, Robert M. 1983. "Political Opinion Polling in Great Britain." In Political Opinion Polling: An International Review, ed. Robert M. Worcester. New York: St. Martins Press. Yalch, Richard F. 1976. "Pre-election Interview Effects on Voter Turnout." Public Opinion Quarterly 40:331-36. Ian McAllister is a professor of politics, the University of New South Wales, Canberra, ACT, Australia 2600. Donley T. Studlar is a professor of political science, Oklahoma State University, Stillwater, OK 74078-0608.